Re: Symlink indirection

[James Antill]

Andrew Walrond <[e-mail]> writes:

> Hi James.
> 
> Thanks for the info; my application is now entirely plausible :)
> And apologies for asking an FAQ. (Google didn't throw up anything useful)

 Sorry, this was an error on my part, I shouldn't reply to Richard's
posts until I've calmed down.

> BTW is documented anywhere except the code?

 Apart from this mailing list, not AFAIK.

> >  The link count is for recursively following symlinks, as the
> > original
> 
> > question wanted to know ... and has been discussed on lkml numerous
> > times.
> >  Andrew, one extra piece of information you might not know is that
> > the
> 
> > above value doesn't come into play when the new symlink is the last
> > element in the new path, then you get a higher value.
> >  The full code...
> >         if (current->link_count >= max_recursive_link)
> 
> >                 goto loop;
> >         if (current->total_link_count >= 40)
> >                 goto loop;
> > [...]
> >         current->link_count++;
> >         current->total_link_count++;
> >         UPDATE_ATIME(dentry->d_inode);
> >         err = dentry->d_inode->i_op->follow_link(dentry, nd);
> >         current->link_count--;
> > ...Ie. a link from /a -> /b/c where "b" is a symlink takes the
> 
> > "max_recursive_link" value (5 on vanilla kernels) but if "/b/c" was a
> > symlink then you get to use the 40 value.

 The last part of the path is still part of the follow_link, and thus
subject to the non-total link_count value. The only times you get the
bigger value is...

/a     -> /b/c
/b     -> /b.1     -> /b.2     -> /b.3     -> /b.4
/b.4/c -> /b.4/c.1 -> /b.4/c.2 -> /b.4/c.3 -> /b.4/c.4

...here you can open /a even if you'll follow more than ->link_count
links.

-- 
# James Antill -- [e-mail]
:0:
* ^From: .*james@and\.org
/dev/null